Research Article

Spin-Torque Switching with the Giant Spin Hall Effect of Tantalum

See allHide authors and affiliations

Science  04 May 2012:
Vol. 336, Issue 6081, pp. 555-558
DOI: 10.1126/science.1218197


Spin currents can apply useful torques in spintronic devices. The spin Hall effect has been proposed as a source of spin current, but its modest strength has limited its usefulness. We report a giant spin Hall effect (SHE) in β-tantalum that generates spin currents intense enough to induce efficient spin-torque switching of ferromagnets at room temperature. We quantify this SHE by three independent methods and demonstrate spin-torque switching of both out-of-plane and in-plane magnetized layers. We furthermore implement a three-terminal device that uses current passing through a tantalum-ferromagnet bilayer to switch a nanomagnet, with a magnetic tunnel junction for read-out. This simple, reliable, and efficient design may eliminate the main obstacles to the development of magnetic memory and nonvolatile spin logic technologies.

Spin-polarized currents can be used to apply torques to magnetic moments by direct transfer of spin angular momentum, enabling manipulation of nanoscale magnetic devices with currents that are orders of magnitude lower than those required for magnetic-field–based control (15). The usual way to generate spin currents strong enough for spin-torque manipulation of magnets has been to send an electron current through a ferromagnetic (FM) polarizing layer, part of a two-terminal magnetic tunnel junction (MTJ) with a layered FM/tunnel barrier/FM structure. Although MTJs can have excellent spin-torque efficiency, they can be challenging to operate reliably; it is difficult to manufacture large-scale memories in which enough spin current can pass through the tunnel barriers to drive reliable magnetic switching without occasionally damaging a barrier. Achieving reliable reading of the MTJ resistance without ever causing switching is also a challenge. It has been known for some time that a spin current can, alternatively, be generated in nonmagnetic materials by the spin Hall effect (SHE) (612), in which spin-orbit coupling causes electrons with different spins to deflect in different directions, yielding a pure spin current transverse to an applied charge current. However, very few attempts have been carried out to use this spin current for manipulating magnetic moments (13, 14).

We report the discovery of a giant SHE in the high-resistivity form of tantalum (β-Ta) (15) and demonstrate that this allows an electrical current in a thin Ta layer to efficiently induce spin-torque switching of an adjacent thin-film ferromagnet, for both perpendicular-to-plane and in-plane magnetized samples at room temperature. High-resistivity β-Ta is produced when Ta is sputter-deposited or evaporated onto amorphous surfaces such as oxidized Si (16) or CoFeB. Ab initio calculations (17) have predicted that highly resistive Ta may have a large spin Hall angle, comparable to or greater than that of Pt and with the opposite sign in comparison to Pt or Au. [Here, the spin Hall angle is defined as θSH = JS/Je, where Je is the charge current density and [ħ/(2e)]JS is the spin current density arising from the SHE (ħ, Planck’s constant h divided by 2π; e, charge on an electron).] In contrast, a nonlocal spin valve measurement (18) reported a very low value for the Ta spin Hall angle, 0.0037 (albeit with the predicted sign). However, this measurement technique can greatly underestimate the spin Hall angle (19, 20), particularly in highly resistive spin Hall materials such as Ta. We have quantified the magnitude of the SHE in Ta using three different methods, and we find the spin Hall angle to be θSHTa=0.12 to 0.15; in comparison with the spin Hall angle of Pt, |θSHPt|0.07 (1924), θSHTa is of opposite sign and larger, whereas it is comparable to one reported value for Pt-doped Au, |θSHAu(Pt)|=0.12±0.04 (25). Unlike Pt, Ta does not substantially increase the magnetic damping (energy dissipation) in an adjacent thin-film magnet; consequently, we show that the giant spin Hall effect spin torque (SHE-ST) from Ta is effective in driving magnetic reversal of in-plane–polarized magnetic layers via an antidamping spin-torque mechanism (5). We use this effect to implement a three-terminal device geometry in which the SHE-ST from Ta produces current-induced switching of the in-plane polarized CoFeB layer, with read-out using a MTJ with a large magnetoresistance. This geometry is straightforward to fabricate and can have comparable efficiency to conventional two-terminal MTJs while providing greatly improved reliability and output signal levels.

Measurement of the SHE in Ta. To determine the spin Hall angle in β-Ta, we used the ST-induced ferromagnetic resonance (ST-FMR) technique, previously introduced in studies of Pt (19). Our samples consisted of Co40Fe40B20(4)/Ta(8) (the numbers in parentheses represent layers; thickness in nanometers) bilayers sputter-deposited onto oxidized Si substrates and patterned into 10-μm-wide strips. We took measurements of the bilayer resistance as a function of varying Ta thickness to determine that the Ta resistivity was ρΤa ≈ 190 microhm·cm, confirming the β-Ta phase. The CoFeB resistivity was ρCoFeB ≈ 170 microhm·cm, and for a 4-nm-thick film, the magnetic moment was oriented in-plane. We applied an oscillating radio frequency current IRF along the strips in the current in-plane configuration, with an external magnetic field Bext in the film plane at a 45° angle with respect to the current direction (Fig. 1A). Because of the SHE, the oscillating current in the Ta generated an oscillating spin current that flowed perpendicular to the sample plane and exerted an oscillating spin torque on the magnetic moment of the CoFeB layer. When the frequency of the bias current and the magnitude of the bias magnetic field satisfied the ferromagnetic resonance condition, magnetic precession occurred. Mixing of IRF and the oscillating anisotropic magnetoresistance of the CoFeB then resulted in a measurable dc voltage. A typical resonance signal is shown in Fig. 1B, where the resonance peak is fitted by the sum of a symmetric Lorentzian and an antisymmetric Lorentzian. The symmetric component of the peak arises from the SHE-ST, whereas the antisymmetric peak is a consequence of the torque generated by the Oersted field from the current in the Ta; the difference in line shape is due to the orthogonal directions of the two torques (Fig. 1A) [for a detailed discussion of the line shapes, see (19, 20)]. We measured the resonant signal for different applied frequencies and found that the positions of the resonant peaks agree well with the Kittel formula f = (γ/2π)[B(B + μ0Meff)]1/2 (Fig. 1B, inset), where γ = 1.76 × 1011 Hz T–1 is the gyromagnetic ratio and μ0Meff is the effective demagnetization field determined to be 1.3 T from the fit.

Fig. 1

ST-FMR induced by the spin Hall effect at room temperature. (A) Sample geometry for the ST-FMR measurement. IRF and HRF represent the applied radio frequency current and the corresponding Oersted field. Embedded Image is the torque on the magnetization due to the Oersted field, and Embedded Image is the spin-transfer torque from the spin Hall effect. Resonant line shapes of the ST-FMR signals under a driving frequency f = 9 GHz for (B) CoFeB(4 nm)/Ta(8 nm) and (C) CoFeB(3 nm)/Pt(6 nm). The squares represent experimental data, whereas the red curves are fits to a sum of symmetric and antisymmetric Lorentzians. From the ratio of the symmetric and antisymmeteric peak components in (C), we determine the JS/Je ratio for Pt to be ~0.07, consistent with earlier work (19). Vmix is the measured dc voltage due to the mixing of oscillating resistance and radio frequency current. The inset to (B) shows the dependence of the frequency f on the resonance magnetic field, in agreement with the Kittel formula (solid curve). (D) The resonance linewidth as determined from ST-FMR signals such as those shown in (B) and (C) at different resonance frequencies. The Gilbert damping coefficients α for Ta and Pt are calculated from the linear fits to these linewidth data. CFB, CoFeB.

To compare the SHE in Ta with that of Pt, we made and measured a different sample with the stack structure: substrate/CoFeB(3)/Pt(6) (thicknesses in nanometers), with the result shown in Fig. 1C. Comparing the resonant signals of CoFeB/Ta and CoFeB/Pt in Fig. 1, B and C, we see that the antisymmetric peaks of the two samples have the same sign, as expected from their common origin (26). The symmetric peaks in the two cases are opposite in sign, which directly shows that the SHE in Ta is opposite to that in Pt, in agreement with the prediction (17) and the previous measurement (18).

We measured the magnitude of the SHE using a self-calibrated technique that uses the ratio of the symmetric peak amplitude S to the antisymmetric peak A to determine the strength of the spin Hall torque relative to the Oersted-field torque (19). Independent of the frequency employed, we found the consistent result that JS/Je = 0.15 ± 0.03 in our 8-nm Ta films. This value of JS/Je represents the spin Hall angle θSH if the spin diffusion length λsf in Ta is much less than the Ta thickness (19, 20). If λsf is comparable to or larger than the film thickness, then the bulk value of θSH is even larger than 0.15 ± 0.03.

If the spin torque from the SHE is to be used for switching nanomagnets by the conventional antidamping ST switching mechanism (5), it is important that the nonmagnetic layer does not substantially increase the effective magnetic damping of the adjacent FM by the spin-pumping effect (27, 28). The ST-FMR measurements discussed above allow a determination of the Gilbert damping coefficient α from the linewidth ΔB (half width at half maximum) of the FMR peak, using the relation α = (γ/2πfB. The results shown in Fig. 1D indicate that α = 0.008 for the CoFeB(4)/Ta(8) bilayer film, close to the intrinsic value expected for a 4-nm-thick CoFeB layer (29) and much smaller than the corresponding α ≈ 0.025 for the CoFeB(3)/Pt(6) sample. This is consistent with (27), in which damping caused by spin pumping was determined to be much stronger in FM/Pt bilayers than in FM/Ta, although the phase of Ta studied in (27) was not reported. Our observation of a strong spin Hall effect in β-Ta is not in conflict with the weakness of the spin-pumping effect in Ta films, because the strength of the spin pumping depends not only on the strength of spin-orbit coupling, but also on the ratio of the elastic scattering time to the spin-flip scattering time and the value of the spin-mixing conductance (28), either or both of which might be smaller in β-Ta than Pt.

Switching a perpendicularly magnetized ferromagnetic layer with the spin Hall effect. Previous experiments (14) using a perpendicularly magnetized FM deposited on Pt, and with a small magnetic field applied in the direction of the electrical current, have demonstrated that the SHE-ST will (once it is strong enough relative to the magnetic anisotropy field) abruptly rotate the out-of-plane moment from the nearly vertical positive (upward) orientation to the nearly vertical negative (downward) orientation, or vice versa, depending on the direction of the current flow and the SHE sign. [Such switching was first reported in (30), but Miron et al. argued that the spin Hall effect was not strong enough to explain their measurements, asserting instead that a Rashba mechanism was dominant.] We have verified that the stronger SHE in Ta can achieve the same switching effect but with the opposite sign compared to Pt. For this measurement, we deposited a thin-film stack with the structure substrate/Ta(4)/Co40Fe40B20(1)/MgO(1.6)/Ta(1) (thicknesses in nanometers) and patterned it into Hall bars 2.5 to 20 μm wide and 3 to 200 μm long (Fig. 2A, inset). MgO was used as a capping layer because previous studies (31) have shown that for a sufficiently thin CoFeB layer, the Ta/CoFeB/MgO structure has a strong perpendicular magnetic anisotropy; this was confirmed by our measurements (Fig. 2A). (The top Ta layer served merely to protect the MgO from exposure to atmosphere.) For the ST switching measurement, a dc current was applied along the strip, and the anomalous Hall resistance RH was recorded to monitor the change in the vertical component of the CoFeB magnetization because RHMz = MSsinθ, where θ represents the angle between the magnetic moment and film plane; Mz is the vertical component of the magnetization, and MS is the saturation magnetization. A static magnetic field Bext was applied almost parallel (or antiparallel) to the in-plane current direction, keeping the angle β between Bext and the film plane fixed, initially at β = 0°. Figure 2B shows an example of the abrupt current-induced switching caused by the SHE-ST, as measured for a 2.5-μm-wide sample with β = 0° and Bext = ±10 mT. The switching curves shown in Fig. 2B are obtained under the same bias conditions as in figure 1 of (14), which reported a similar effect for the Pt/Co/AlOx system. Comparison between the two reveals that the switching direction caused by the in-plane current in Fig. 2B is opposite to that in the Pt/Co/AlOx system. We made additional control samples from a Pt/CoFeB/MgO multilayer and found that the switching direction is the same as with Pt/Co/AlOx (fig. S5) (32), demonstrating that the sign reversal comes from the difference between the sign of the SHE in Pt and Ta, and not from any differences between the FM/oxide interfaces or between Co and CoFeB.

Fig. 2

Spin Hall effect–induced magnetic switching in a perpendicularly magnetized Ta/CoFeB/MgO/Ta film at room temperature. (A) The anomalous Hall resistance RH as a function of magnetic field when Bext is applied along the easy axis (perpendicular to the film plane). (Inset) Device geometry used for the measurement. Bext is applied in the plane defined by the direction of current flow and the normal vector to the sample plane. β is the angle between the direction of Bext and the applied current. (B) Current-induced switching when Bext is parallel (top) or antiparallel (bottom) to the current direction defined as in the inset to (A). In both panels, β = 0°. (C) RH versus Bext determined experimentally when the field is applied at the angle β = 2°. Constant currents of ±0.7 mA were applied to the sample while sweeping the field. (D) ΔB(RH) as determined from the difference of the two data sets in (C). The green curve is a fit to the macrospin model. (Inset) Values of Embedded Image determined at different bias currents.

To quantitatively determine the magnitude of the spin Hall angle from the response of perpendicularly magnetized Ta/CoFeB/MgO samples, we swept the magnetic field, keeping its direction at a small field angle β ≈ 2°. With a nonzero β, the vertical component of the external magnetic field Bz = Bext sinβ causes the magnetization of the Hall bar structure to remain uniformly magnetized as long as the current is well below the switching point, so that the magnetization rotates coherently with field and current (Fig. 2C). For convenience in the data analysis, we will treat Bext as function of RH instead of the reverse. As demonstrated in (14), the difference between the Bext(RH) curves for I = +0.7 mA and –0.7 mA can be shown, within a macrospin model, to be proportional to the applied spin torque: ΔB[RH(θ)]=B+(θ)B(θ)=2τST0/sin(θβ). Here, B+/–(θ) is defined as the value of Bext required to produce a given value of the magnetization angle θ when I is positive/negative. Figure 2D shows ΔB(RH) determined by subtracting the two data sets in Fig. 2C. We plot RH normalized with respect to its maximum value, so that it is equal to sinθ. Using a one-parameter fit, the magnitude of the spin torque can be determined to be τST02.1 mT for |I| = 0.7 mA. The τST0/I ratios obtained for different values of applied current are summarized in the inset of Fig. 2D; on average, we find τST0/I2.8±0.6 mT/mA. By using the formula JS=2eMStτST0/

with saturation magnetization MS = (1.1 ± 0.2) × 106 A/m and CoFeB film thickness t = 1.0 ± 0.1 nm, we obtain JS/Je = 0.12 ± 0.03 for the 4-nm Ta layer, consistent with the value JS/Je = 0.15 ± 0.03 from the ST-FMR study for an 8-nm Ta layer. Here, we assume a uniform current density throughout both the Ta and CoFeB layers, because their resistivities are similar: ρTa ≈ 190 microhm·cmm and ρCoFeB ≈ 170 microhm·cm (32).

Spin-torque switching of an in-plane polarized magnet using a three-terminal spin Hall device. The giant SHE in Ta, together with its small effect on the damping of adjacent magnetic layers, makes Ta an excellent material for effecting ST switching of an in-plane magnetized nanomagnet. In conventional antidamping ST switching where the spins are injected either nearly parallel or antiparallel to the initial orientation of the local magnetic moment, the critical current density for switching in the absence of thermal fluctuations is given by (3, 33)JC02eμ0MStα(HC+Meff/2)/(JS/Je) (1)where HC represents the coercive field of the FM nanomagnet.

To demonstrate in-plane magnetic switching induced by the SHE, we fabricated a three-terminal device, consisting of the multilayer substrate/Ta(6.2)/Co40Fe40B20(1.6)/MgO(1.6)/ CoFeB(3.8)/Ta(5)/Ru(5) (thicknesses in nanometers) patterned (32) into the geometry shown in Fig. 3A. The Ta bottom layer was patterned into a 1-μm-wide, 5-μm-long strip (with resistance 3 kilohm), and the rest of the layers were etched to form a MTJ nanopillar on top of the Ta with lateral dimensions ~100 by 350 nm and with the long axis of the nanopillar perpendicular to the long axis of the Ta microstrip.

Fig. 3

Spin Hall effect–induced switching for an in-plane magnetized nanomagnet at room temperature. (A) Schematic of the three-terminal SHE devices and the circuit for measurements. The direction of the spin Hall spin transfer torque is not the same as in Fig. 1A because the CoFeB layer now lies above the Ta rather than below. (B) TMR minor loop of the MTJ as a function of the external applied field Bext applied in-plane along the long axis of the sample. (Inset) TMR major loop of the device. (C) TMR of the device as a function of applied dc current IDC. An in-plane external field of –3.5 mT is applied to set the device at the center of the minor loop. (D) Switching currents as a function of the ramp rate for sweeping current. Red squares indicate switching from AP to P; blue triangles indicate switching from P to AP. Solid lines represent linear fits of switching current versus log(ramp rate). Error bars are smaller than the symbol size.

The magnetoresistance response of one of these MTJs is shown in Fig. 3B, which indicates a coercive field BC ≈ 4 mT, a zero-bias MTJ resistance RMTJ ≈ 65 kilohm, and a tunneling magnetoresistance (TMR) ≈ 50%. During subsequent magnetic switching measurements, we applied a –3.5-mT in-plane magnetic field along the long axis of the MTJ to cancel the dipole field from the top layer of the MTJ acting on the bottom layer and, thus, biased the junction at the midpoint of its minor magnetoresistance loop. We then applied a dc current ITa to the Ta microstrip while monitoring the differential resistance dV/dI of the MTJ (Fig. 3A). Figure 3C shows that abrupt hysteretic switching of the MTJ resistance occurred when ITa was swept through 1 mA, which resulted in antiparallel to parallel (AP-P) switching, and then this switching was reversed (P-AP switching) when the current was swept back past –1 mA.

We have considered other potential mechanisms for this switching besides the SHE-ST. The Oersted field generated by the current can be ruled out because it has the polarity to oppose the switching that we observe, and it is small [0.7 mT at 1 mA (32)] relative to the coercive field. We can also rule out the effect of any in-plane Rashba field (30, 34) that might be generated by ITa, because we measured the switching phase diagram of our three-terminal devices as the function of current and applied in-plane magnetic field (fig. S4) (32). The result is as expected for thermally assisted antidamping ST switching (35) and is inconsistent with switching resulting from any type of current-generated effective field. In addition, the direction that has been reported (36) for the in-plane Rashba field in a Ta/CoFeB/MgO multilayer, if it exists, is the same as the Oersted field and would therefore also act to oppose the switching that we observe. We conclude that the switching we measure is indeed the result of the ST exerted on the bottom MTJ electrode by the transverse spin current from the giant SHE in Ta.

By varying the current ramp rate (Fig. 3D) and using the standard model for thermally activated ST switching (35), we determined both the zero–thermal-fluctuation ST critical currents and the energy barriers for the thermally activated AP-P and P-AP transitions. We found the two critical currents (|Ic0| = 2.0 ± 0.1 mA) and energy barriers [U = 45.7 ± 0.5kBT (kB, the Boltzmann constant; T, temperature)] to be essentially the same. The latter is not surprising, but the former, although consistent with a SHE origin, is distinctly different from the case for ST switching by the spin-polarized current generated by spin filtering within a spin valve or MTJ, where, in general, |Ic0.P-AP| ≠ |Ic0.AP-P| due to spin accumulation in the spin valve and the MTJ magnetoresistance behavior, respectively. The equivalence of the two critical currents for a SHE-ST switching device could be a major technical advantage. From our measured values of |Ic0| and using Eq. 1 with μ0Meff = 0.76 T (32), we determine JS/Je for this device to be 0.12 ± 0.04 (32), in accord with our two other spin Hall angle measurements. We note that our three determinations of JS/Je are consistent for FM layer thicknesses ranging from 1 to 4 nm and are not sensitive to whether the FM layer is magnetized in plane or out of plane.

Technology applications. Improvements to this initial three-terminal SHE device can be very reasonably expected to result in substantial reductions in the switching currents for thermally stable nanomagnets. By reducing the width of the Ta microstrip to be equal to the dimension of the long axis of the nanopillar, we can easily decrease Ic0 by a factor of 3 without affecting thermal stability. A further reduction in Ic0 could be achieved by reducing the demagnetization field of the FM free layer from 700 mT to ≤100 mT (37, 38). With such improvements, Ic0 could be reduced to <100 μA, at which point the three-terminal SHE devices would be competitive with the efficiency of conventional ST switching in optimized MTJs (31, 33, 39) while providing the added advantage of a separation between the low-impedance switching (write) process and high-impedance sensing (read) process. This separation solves the reliability challenges that presently limit applications based on conventional two-terminal MTJs while also giving improved output signals. Other three-terminal spin-torque devices based on conventional spin-filtering have been demonstrated previously (4043), but the SHE-ST design can provide better spin-torque efficiency and is much easier to fabricate. Moreover, the discovery of materials with even larger values of the spin Hall angle than in β-Ta could also add to the competitiveness of the SHE-ST.

Supplementary Materials

Materials and Methods

Supplementary Text

Figs. S1 to S5

Reference (44)

References and Notes

  1. The antisymmetric peaks shown here for these two samples are opposite to what was illustrated in (20) for a substrate/Pt/Permalloy sample, because the relative order of the FM/nonmagnetic layers was reversed in that case.
  2. Supplementary materials are available on Science Online.
  3. Acknowledgments: We acknowledge support from the Army Research Office, Defense Advanced Research Projects Agency, Office of Naval Research, and NSF/Materials Research Science and Engineering Center (DMR-1120296) through the Cornell Center for Materials Research (CCMR), as well as the NSF/Nanoscale Science and Engineering Center Program through the Cornell Center for Nanoscale Systems. We also acknowledge NSF support through use of the Cornell Nanofabrication Facility/National Nanofabrication Infrastructure Network and the CCMR facilities. Patent disclosures have been filed on behalf of the authors regarding the use of the spin Hall effect in Ta for magnetic memory and logic applications.
View Abstract

Navigate This Article